Introduction/MotivationProposed variants

Experimental results

A PSO-Based Mobile Robot for Odor SourceLocalization in Dynamic Advection-Di�usion withObstacles Environment: Theory, Simulation and

Measurement

Wisnu Jatmiko,Kosuke Sekiyama and Toshio Fukuda

IEEE Computational Intelligence Magazine - 2007 (JCR: 2.833)

February 8, 2012

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Paper outline

1 Introduction/Motivation2 PSO framework

1 Standard PSO2 Proposal #1: Detection and Responding PSO (DR-PSO)3 Proposal #2: Charged PSO (C-PSO)

3 Implementation Framework1 Environment2 Robot behavior3 Obstacle-free experiments4 Obstacle-�lled experiments

4 Extension with Wind1 Odor-gated rheotaxis (OGR): particles use not only odor

particle concentration but also wind direction2 Implementation I: used forbidden area3 Implementation II: xθ parameter

5 ConclusionsBorja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Goal

Odor source localization using a robot swarm - each robot ismodeled as a particle

Standard PSO is not appropriate to this problem because itdoesn't react to changing environments

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Standard PSO

Standard PSO equations:

,

where χ is a constriction factor, c1 and c2 are learning parameters,such that c1 = c2 = 2 (????). pi represents the best local data andpg the best global (assumed to be shared among robots/particles)

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

PSO: Non-stationarity problem

Figure: PSO and changing environments

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Detection and Responding PSO I

Standard PSO but, whenever a change is detected, particlesspread randomly for a �xed amount of time

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Detection and Responding PSO II

Figure: DR-PSO response to changes in environment

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Detection and Responding PSO III

Figure: DR-PSO response to changes in environment

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Charged PSO I

Coulomb law is added to the particle dynamic model to keepdiversity of the position of the particles

Some particles are charged, some are not

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Charged PSO II

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Common settings

Robots �rst search for a trace of odor, then PSO drive themtoward the source and �nally the source is declared

The Advection-Di�usion odor model is used (Farrell et al.) togenerate a dynamic plume

Odor and position sensors model random noise

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Obstacle-free I

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Obstacle-free II

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Obstacle-�lled I

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Obstacle-�lled II

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Odor-gated Rheotaxis (OGR)

Robots are able to perceive odor-particle concentration andwind direction

Bio-inspired technique

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Implementation I: Use forbidden area

,

where θ is the angle between the particle's current velocity andwind direction

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Implementation I: diagram I

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Implementation I: diagram II

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Implementation II: χθ parameter

Instead of avoiding movements within forbidden area,velocities are weigthed with a new parameter: χθ

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Implementation II: diagram

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

C-PSO vs. Wind-use Implementation I & II

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement

institution-logo

Introduction/MotivationProposed variants

Experimental results

Using only odor particle concentrationUsing odor particle concentration and wind direction

Wind-use Implementation I vs. II

Borja F.G. A PSO-Based Mobile Robot for Odor Source Localization in Dynamic Advection-Di�usion with Obstacles Environment: Theory, Simulation and Measurement